app.py

import streamlit as st
from transformers import pipeline
import re

st.set_page_config(page_title="Hindi Sentiment Analysis", layout="centered")

# Custom CSS for styling
st.markdown("""
    <style>
    @import url('https://fonts.googleapis.com/css2?family=Poppins&display=swap');

    html, body, [class*="css"] {
        font-family: 'Poppins', sans-serif;
        text-align: center;
    }

    .centered-title {
        font-size: 50px;
        font-weight: bold;
        margin-bottom: 10px;
    }

    .sentiment-result {
        font-size: 26px;
        font-weight: 600;
        margin-top: 20px;
    }

    .confidence {
        font-size: 20px;
        margin-top: 10px;
    }

    .original-text {
        font-style: italic;
        margin-top: 10px;
    }

    textarea {
        text-align: left !important;
    }
    </style>
""", unsafe_allow_html=True)

# Load model
pipe = pipeline("text-classification", model="NeonSamurai/hindi_sentiment_bert_finetuned")

names = ["neutral", "positive", "negative"]
emojis = {"positive": "🤗", "negative": "😔", "neutral": "😐"}

# Utility functions
def is_mostly_hindi(text):
    if not text.strip():
        return False
    devanagari_pattern = r'[\u0900-\u097F]'
    allowed_pattern = r'[a-zA-Z0-9\s.,!?]'
    devanagari_chars = len(re.findall(devanagari_pattern, text))
    allowed_chars = len(re.findall(allowed_pattern, text))
    total_chars = len(text)
    hindi_proportion = devanagari_chars / total_chars if total_chars > 0 else 0
    valid_chars = devanagari_chars + allowed_chars == total_chars
    return hindi_proportion >= 0.7 and valid_chars

def clean_input(text):
    cleaned_text = re.sub(r'[^a-zA-Z0-9\u0900-\u097F\s?.!]', ' ', text)
    cleaned_text = re.sub(r'([?.!])(?![?.!]\s|$)', '', cleaned_text)
    cleaned_text = ' '.join(cleaned_text.split())
    return cleaned_text

# Title
st.markdown("<div class='centered-title'>📊 Hindi Sentiment Analysis</div>", unsafe_allow_html=True)

# Instructions
with st.expander("ℹ️ Instructions"):
    st.markdown("""
    - Please enter a sentence or paragraph in **Hindi (Devanagari script)**.
    - Example: *यह फिल्म बहुत अच्छी थी और अभिनय शानदार था।*
    - The app will classify the text as **positive**, **negative**, or **neutral**.
    """)

# Text input
user_input = st.text_area("✍️ Enter Hindi text below:", height=150, placeholder="Type your Hindi text here...")

# Predict button
if st.button("🔍 Predict Sentiment"):
    if not user_input.strip():
        st.warning("⚠️ Please enter some text.")
    else:
        cleaned_input = clean_input(user_input)
        if not is_mostly_hindi(cleaned_input):
            st.error("❌ Input should be primarily in Hindi (Devanagari script).")
        else:
            result = int(pipe(cleaned_input)[0]['label'].split("_")[1])
            sentiment = names[result]
            emoji = emojis[sentiment]

            st.markdown(f"<div class='sentiment-result'>✅ Sentiment: <b>{sentiment.capitalize()} {emoji}</b></div>", unsafe_allow_html=True)
            st.markdown(f"<div class='confidence'>Confidence Score: <code>{result['score']:.2f}</code></div>", unsafe_allow_html=True)
            st.markdown(f"<div class='original-text'>“{user_input}”</div>", unsafe_allow_html=True)

# Background-Image
st.markdown(
    """
    <style>
        .stApp {
            background-image: url("https://cdn-uploads.huggingface.co/production/uploads/673f5e166c2774fcc8a82f0b/OIaLFHTZGXeTvsizCCNsK.png");
            background-size: cover;
            background-position: center;
            height: 100vh;
        }
        
        /* Semi-transparent overlay */
        .stApp::before {
            content: "";
            position: absolute;
            top: 0;
            left: 0;
            width: 100%;
            height: 100%;
            background: rgba(0, 0, 0, 0.4);  /* 40% transparency */
            z-index: -1;
        }
    </style>
    """, 
    unsafe_allow_html=True
)